AI-Based Graph Neural Networks for Predictive Energy Optimization in IoT-Enabled Consumer Electronics

This study presents an innovative framework leveraging Graph Neural Networks (GNNs) integrated with reinforcement learning (RL) and meta-learning for predictive energy optimization in IoT-enabled consumer electronics. The proposed methodology addresses the challenges of dynamic energy demands, sparse datasets, and real-time adaptability by combining the decision-making capabilities of RL with the rapid adaptability of meta-learning. RL-GNNs employ Proximal Policy Optimization (PPO) to dynamically schedule energy usage, balancing exploration and exploitation to optimize energy consumption. Concurrently, meta-learning-enhanced GNNs (Meta-GNNs) utilize Model-Agnostic Meta-Learning (MAML) for few-shot learning, enabling efficient adaptation to evolving IoT environments. Extensive experiments on simulated and real-world IoT datasets demonstrate the framework’s superiority over conventional models, achieving up to 25.2% energy savings and over 92% prediction accuracy. Results highlight significant improvements in computational efficiency and adaptability to sparse and noisy data scenarios, confirming the framework’s versatility. This work establishes a novel approach to IoT energy optimization, offering transformative potential for applications in smart grids, industrial IoT, and connected home systems. The integration of RL and meta-learning into GNN architectures marks a significant advancement in energy-efficient IoT system design.